ARGIRA v1.3.0-alpha: Three Layer Uncertainty Communication Prototype for Automatic Alt-Text Accessibility Pilot Release

ARGIRA v1.3.0-alpha

ARGIRA is a research prototype that generates alternative text for images using a lightweight statistical heuristic together with a three-layer architecture for communicating the instrument's own uncertainty.

The guiding principle is that uncertainty messages describe the behavior of the instrument, not the truth of the image itself.

ARGIRA does not claim whether a classification is correct. Instead, it communicates what is known about the instrument's historical performance when producing similar outputs.

Architecture

ARGIRA consists of three main layers.

Layer 1 — Image analysis and alternative text generation

Layer 1 classifies the image type:

photo

painting

illustration

possibly synthetic

It uses weighted image statistics and generates the proposed alternative text.

Layer 2 — Historical behavior explanation

Layer 2 provides, on request, a plain-language explanation describing how the instrument has historically behaved when producing similar outputs.

This information is based on a characterization corpus of 162 images.

Layer 3 — Technical inspection and auditing

Layer 3 provides technical information, including:

confidence

margin

calibration state

supporting metadata

This layer is intended for inspection, auditing, and research.

Version history

Version 1.2

Version 1.2 extended the demonstration and result-management infrastructure while preserving the validated analytical pipeline from v1.1.

The heuristic (Layers 1–2.5) remained unchanged.

This release introduced Layer 3.5 (Result Manager), providing:

in-memory result persistence

multi-format export (TXT, CSV, and JSON)

a single serialization interface

clipboard copy

file download for supported formats

accessibility improvements, including:

a skip link

explicit ARIA relationships between disclosure controls and associated panels

Version 1.3.0-alpha

Version 1.3.0-alpha extends this infrastructure toward exploratory accessibility evaluation.

The analytical pipeline remains unchanged from v1.2.

This release introduces:

session-level instrumentation

event-level instrumentation

session identifiers attached to results

lightweight recording of interactions with uncertainty disclosure controls

This infrastructure is intended to support future accessibility evaluation with screen reader users.

No user study or comparative analysis has been performed in this release.

Analytical equivalence

For the same input image:

image classification does not differ between v1.2 and v1.3.0-alpha

calibration assessment does not differ

confidence estimation does not differ

generated alternative text does not differ

Research status

ARGIRA v1.3.0-alpha is released as a research prototype for documentation and experimentation.

It is not intended to be used as a validated production image classification system.

The Layer 2 and Layer 3 validation performed for v1.1 was not repeated for v1.3.0-alpha because the validated analytical pipeline remained unchanged.

The following components are infrastructure components rather than analytical components:

Result Manager (Layer 3.5)

session instrumentation

event instrumentation

These components have not themselves undergone a formal validation study.